Patient-specific detection of ventricular tachycardia in remote continuous health devices

Ventricular Tachycardia (VT) is a dangerous arrhythmic event which can lead to sudden cardiac death if not detected and taken care of in time. This work uses non-linear features derived from Recurrence Quantification Analysis (RQA) along with Kolmogorov complexity, by analyzing the ECG signals, to train a classifier which can predict VT prior to their onset in remote continuous health devices. Compressed ECG signal along with amplitude ranges extracted from the ECG signal are used as features to strengthen the classifier. Stacked Denoising Autoencoders (SDAE) are used for the purpose of feature extraction and compression of signals, and their performance is compared with other works that detect VT for different window sizes. Softmax Regression is used as the classifier in this work. The proposed method is tested against MIT-BIH Arrhythmia database, MIT-BIH Malignant Ventricular Arrhythmia Database (VFDB) and Creighton University Ventricular Tachyarrhythmia Database (CUDB). A total of 96.52% accuracy with 96.18% sensitivity is obtained after testing the proposed method on all test records.